ABSTRACT
Coastal bridges are susceptible to extreme waves during coastal natural hazards like tropical cyclones and tsunamis. The investigation of the structural performance of coastal bridges under extreme wave impacts plays a central role in the development of coastal communities. In recent years, deep neural networks based methods emerge and shown promising performance for these hydrodynamic problems. However, their interpretability and generalization remain a major challenge, especially for the complex uncertainties within the hydrodynamic process. In this study, the authors propose to combine the advantages of both experimental studies and deep neural networks for such investigations, so as to achieve reliability, network interpretability, and effectiveness simultaneously. Specifically, laboratory experiments are conducted to test the wave impacts on coastal bridges under various wave scenarios and hydrological environments. The measured wave profiles and wave loads are divided into a multivariate time series and used as the input of a Long-Short-Term-Memory (LSTM) model. Such a prediction method helps to reduce errors caused by experimental uncertainty by substituting every set of the collected data during the repeated experiments into the training set. In summary, the proposed scheme is effective and interpretable, and the comparisons with experimental results demonstrate its superior performance in predicting the structural performance of coastal bridges under wave impacts.
